Estimation of heart rate power spectrum bands from real-world data: dealing with ectopic beats and noisy data

The authors examine two different methods for characterizing the heart rate (HR) power spectrum of data with bad intervals containing ectopic or missing beats. Both methods estimate the spectrum using the fast Fourier transform of the HR autocorrelation. Method I computes the autocorrelation without making explicit assumptions about the specific value of the HR in the bad intervals. The only assumption is that the bad intervals are relatively independent of the HR. Method II uses linear splines to fill in the HR during bad intervals. Comparing the performance of the methods for the specific problem of determining the low (0.02-0.1 Hz) and high (0.15-0.4 Hz) frequency powers, it is found that method II performs substantially better than method I. This finding, however, applies only to the specific spectral measures tested, and for other measures method I may yield better results